Classifying views of an angiographic medical imaging system

Abstract

Systems and methods are provided for acquiring a series of angiographic images and identifying the anatomical structures represented in the series of images using a machine learnt classifier. Additional series of images that would yield the optimal visualization of the structure of interest may be suggested.

We claim: 
     
       1. A method for classifying a video in coronary angiography, the method comprising:
 acquiring a temporal series of angiographic image frames; 
 removing one or more angiographic image frames from the temporal series of angiographic image frames; 
 classifying by a machine-learnt classifier each of the remaining angiographic image frames of the temporal series of angiographic image frames as visualizing a first anatomical structure or a second anatomical structure; 
 labeling the temporal series of angiographic image frames as visualizing the first anatomical structure or the second anatomical structure based on the classifications of the remaining images; and 
 providing the label of the temporal series of angiographic image frames to a user. 
 
     
     
       2. The method of  claim 1 , wherein the first anatomical structure is a left coronary artery and the second anatomical structure is a right coronary artery. 
     
     
       3. The method of  claim 1 , wherein acquiring the temporal series of angiographic image frames is performed by an x-ray imaging device. 
     
     
       4. The method of  claim 1 , wherein the one or more angiographic image frames are removed based on a contrast level. 
     
     
       5. The method of  claim 4 , wherein removing comprises:
 denoising an image frame of the angiographic image frames; 
 generating a binary image of the denoised image frame; and 
 removing the binary denoised image frame when a contrast level does not reach a predefined threshold. 
 
     
     
       6. The method of  claim 1 , wherein removing comprises:
 removing a first percentage of angiographic image frames from a start of the temporal series and a second percentage of angiographic image frames from an end of the temporal series. 
 
     
     
       7. The method of  claim 1 , wherein classifying by a machine learnt classifier comprises:
 training a convolutional neural network with training data comprising labeled prior image frames; 
 inputting each of the remaining images frames into the convolutional neural network; and 
 returning a classification from the convolutional neural network. 
 
     
     
       8. The method of  claim 1 , wherein labeling comprises:
 comparing a number of angiographic image frames classified as the first anatomical structure versus the second anatomical structure; and 
 selecting the first anatomical structure or the second anatomical structure based on the comparison. 
 
     
     
       9. The method of  claim 1 , further comprising:
 generating the optimality score for the temporal series of angiographic image frames as a function of:
     s =(( a−n/ 2)*2/ n )2 wherein: 
 (a)=a number of angiographic frames classified as the first anatomical structure; 
 (n)=a total number of angiographic frames; 
 (s)=the optimality score; and 
 
 providing the optimality score to the user. 
 
     
     
       10. A method for classifying views in angiography, the method comprising:
 acquiring a temporal series of angiographic images; 
 classifying, using a deep machine-learnt classifier, a view of an anatomical structure in each of the angiographic images of the temporal series of angiographic images; 
 scoring the temporal series of angiographic images based on the classifications of the view in each of the angiographic images; 
 identifying, based on the score, a second view of the anatomical structure, wherein the second view comprises one or more scan parameters for acquiring the second view; and 
 acquiring a second set of angiographic images with the one or more scan parameters. 
 
     
     
       11. The method of  claim 10 , wherein the view is classified as either a view of a left coronary artery or a view of a right coronary artery. 
     
     
       12. The method of  claim 11 , wherein scoring comprises:
 generating the score for the temporal series as a function of:
     s =(( a−n/ 2)*2/ n )2 wherein: 
 (a)=a number of angiographic images classified as the view of the left coronary artery; 
 (n)=a total number of angiographic images; and 
 (s)=the score. 
 
 
     
     
       13. The method of  claim 10 , further comprising:
 removing, prior to classifying, one or more angiographic images from the temporal series of angiographic images as a function of a contrast level in the one or more angiographic images. 
 
     
     
       14. The method of  claim 10 , further comprising:
 identifying, when acquiring the temporal series, an angle of a C-arm; wherein identifying the second view is further based on the angle. 
 
     
     
       15. A system for classifying views in angiography, the system comprising:
 a medical image scanner configured to acquire a temporal series of angiographic images; 
 a processor configured to classify each image of the temporal series of angiographic images as visualizing an anatomical structure using a machine-learnt classifier, the processor further configured to determine a label for the temporal series of angiographic images based on the classifications of each of the images; and 
 a display configured to display the temporal series of angiographic images and the label. 
 
     
     
       16. The system of  claim 15 , wherein each of the images are classified as either a view of a left coronary artery or a view of a right coronary artery. 
     
     
       17. The system of  claim 15 , wherein the processor is configured to classify each of the images using a convolutional neural network. 
     
     
       18. The system of  claim 17 , wherein the convolutional neural network is trained using image data and classification data from prior labeled images. 
     
     
       19. The system of  claim 17 , wherein the processor is configured to classify each of the images using two or more convolutional neural networks. 
     
     
       20. The system of  claim 15 , wherein the processor is further configured to calculate an optimality score for the temporal series based on the classifications.